def download_commits(repo_dir):
hg = hglib.open(repo_dir)
commits = hg.log()
bug_pattern = re.compile('[\t ]*[Bb][Uu][Gg][\t ]*([0-9]+)')
def transform(commit):
desc = commit[5].decode('utf-8')
bug_id = None
bug_id_match = re.search(bug_pattern, desc)
if bug_id_match:
bug_id = int(bug_id_match.group(1))
return {
# 'rev': commit[0].decode('utf-8'),
# 'node': commit[1].decode('utf-8'),
# 'tags': commit[2].decode('utf-8'),
# 'branch': commit[3].decode('utf-8'),
# 'author': commit[4].decode('utf-8'),
'desc': desc,
# 'date': str(commit[6]),
'bug_id': bug_id,
}
commits = [transform(commit) for commit in reversed(commits)]
db.write(COMMITS_DB, commits)
def get_commits_to_ignore(self):
logger.info("Download previous commits to ignore...")
db.download(IGNORED_COMMITS_DB)
logger.info("Get previously classified commits...")
prev_commits_to_ignore = list(db.read(IGNORED_COMMITS_DB))
logger.info(
f"Already found {len(prev_commits_to_ignore)} commits to ignore..."
)
# When we already have some analyzed commits, re-analyze the last 3500 to make sure
# we didn't miss back-outs that happened since the last analysis.
if len(prev_commits_to_ignore) > 0:
first_commit_to_reanalyze = (
-3500 if len(prev_commits_to_ignore) >= 3500 else 0)
rev_start = "children({})".format(
prev_commits_to_ignore[first_commit_to_reanalyze]["rev"])
else:
rev_start = 0
with hglib.open(self.mercurial_repo_dir) as hg:
revs = repository.get_revs(hg, rev_start)
commits = repository.hg_log_multi(self.mercurial_repo_dir, revs)
with hglib.open(self.mercurial_repo_dir) as hg:
repository.set_commits_to_ignore(hg, self.mercurial_repo_dir,
commits)
for commit in commits:
commit.ignored |= commit.author_email == "*****@*****.**"
chosen_commits = set()
commits_to_ignore = []
for commit in commits:
if commit.ignored or commit.backedoutby:
commits_to_ignore.append({
"rev":
commit.node,
"type":
"backedout" if commit.backedoutby else "",
})
chosen_commits.add(commit.node)
logger.info(f"{len(commits_to_ignore)} new commits to ignore...")
for prev_commit in prev_commits_to_ignore[::-1]:
if prev_commit["rev"] not in chosen_commits:
commits_to_ignore.append(prev_commit)
chosen_commits.add(prev_commit["rev"])
logger.info(f"{len(commits_to_ignore)} commits to ignore...")
logger.info("...of which {} are backed-out".format(
sum(1 for commit in commits_to_ignore
if commit["type"] == "backedout")))
db.write(IGNORED_COMMITS_DB, commits_to_ignore)
zstd_compress(IGNORED_COMMITS_DB)
db.upload(IGNORED_COMMITS_DB)
def test_write_read(tmp_path):
db_path = tmp_path / 'prova.json'
db.register(db_path, 'https://alink', 1)
db.write(db_path, range(1, 8))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7]
def get_commits_to_ignore(self) -> None:
assert db.download(repository.COMMITS_DB)
ignored = set()
commits_to_ignore = []
all_commits = set()
annotate_ignore_nodes = {
node
for node, label in labels.get_labels("annotateignore")
if label == "1"
}
for commit in repository.get_commits(include_no_bug=True,
include_backouts=True,
include_ignored=True):
all_commits.add(commit["node"][:12])
if (commit["ignored"] or commit["backedoutby"]
or not commit["bug_id"] or len(commit["backsout"]) > 0
or repository.is_wptsync(commit)
or commit["node"] in annotate_ignore_nodes):
commits_to_ignore.append({
"rev":
commit["node"],
"type":
"backedout" if commit["backedoutby"] else "",
})
ignored.add(commit["node"][:12])
if len(commit["backsout"]) > 0:
for backedout in commit["backsout"]:
if backedout[:12] in ignored:
continue
ignored.add(backedout[:12])
commits_to_ignore.append({
"rev": backedout,
"type": "backedout"
})
logger.info(f"{len(commits_to_ignore)} commits to ignore...")
# Skip backed-out commits which aren't in the repository (commits which landed *before* the Mercurial history
# started, and backouts which mentioned a bad hash in their message).
commits_to_ignore = [
c for c in commits_to_ignore if c["rev"][:12] in all_commits
]
logger.info(f"{len(commits_to_ignore)} commits to ignore...")
logger.info("...of which {} are backed-out".format(
sum(1 for commit in commits_to_ignore
if commit["type"] == "backedout")))
db.write(IGNORED_COMMITS_DB, commits_to_ignore)
zstd_compress(IGNORED_COMMITS_DB)
db.upload(IGNORED_COMMITS_DB)
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
first_rev = get_rev(hg, date_from)
commits = hg_log(hg, first_rev)
commits_num = len(commits)
hg.close()
# Total previous number of commits by the author.
total_commits_by_author = defaultdict(int)
# Previous commits by the author, in a 90 days window.
commits_by_author = defaultdict(list)
global author_experience
global author_experience_90_days
for commit in commits:
author_experience[commit] = total_commits_by_author[commit.author]
total_commits_by_author[commit.author] += 1
# Keep only the previous commits from a window of 90 days in the commits_by_author map.
cut = None
for i, prev_commit in enumerate(commits_by_author[commit.author]):
if (commit.date - prev_commit.date).days <= 90:
break
cut = i
if cut is not None:
commits_by_author[commit.author] = commits_by_author[
commit.author][cut + 1:]
author_experience_90_days[commit] = len(
commits_by_author[commit.author])
commits_by_author[commit.author].append(commit)
subprocess.run([
os.path.join(repo_dir, 'mach'), 'file-info', 'bugzilla-automation',
'component_data'
],
cwd=repo_dir,
check=True)
global COMPONENTS
with open(os.path.join(repo_dir, 'component_data',
'components.json')) as cf:
COMPONENTS = json.load(cf)
print(f'Mining commits using {multiprocessing.cpu_count()} processes...')
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def compress_and_upload() -> None:
db.write(
SHADOW_SCHEDULER_STATS_DB,
(scheduler_stats[push.rev]
for push in pushes if push.rev in scheduler_stats),
)
utils.zstd_compress(SHADOW_SCHEDULER_STATS_DB)
db.upload(SHADOW_SCHEDULER_STATS_DB)
def test_bad_format_compression(tmp_path, db_name):
db_path = tmp_path / db_name
db.register(db_path, "https://alink")
with pytest.raises(AssertionError):
db.write(db_path, range(7))
with pytest.raises(AssertionError):
db.append(db_path, range(7))
def retrieve_test_info(self, days: int) -> Dict[str, Any]:
logger.info("Download previous test info...")
db.download(TEST_INFOS_DB)
dates = [
datetime.utcnow() - timedelta(days=day)
for day in reversed(range(days))
]
logger.info("Get previously gathered test info...")
test_infos = {
test_info["date"]: test_info
for test_info in db.read(TEST_INFOS_DB)
}
prev_skips = None
for date in tqdm(dates):
date_str = date.strftime("%Y-%m-%d")
# Gather the latest three days again, as the data might have changed.
if date_str in test_infos and date < datetime.utcnow() - timedelta(
days=3):
prev_skips = test_infos[date_str]["skips"]
continue
test_infos[date_str] = {
"date":
date_str,
"bugs": [{
"id": item["bug_id"],
"count": item["bug_count"]
} for item in test_scheduling.get_failure_bugs(date, date)],
"skips": {},
}
try:
test_info = test_scheduling.get_test_info(date)
for component in test_info["tests"].keys():
test_infos[date_str]["skips"][component] = sum(
1 for test in test_info["tests"][component]
if "skip-if" in test)
except requests.exceptions.HTTPError:
# If we couldn't find a test info artifact for the given date, assume the number of skip-ifs didn't change from the previous day.
assert prev_skips is not None
test_infos[date_str]["skips"] = prev_skips
prev_skips = test_infos[date_str]["skips"]
db.write(
TEST_INFOS_DB,
(test_infos[date.strftime("%Y-%m-%d")]
for date in dates if date.strftime("%Y-%m-%d") in test_infos),
)
zstd_compress(TEST_INFOS_DB)
return test_infos
def download_commits(repo_dir):
commits = hg_log(repo_dir)
commits_num = len(commits)
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=256)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def test_bad_format_compression(tmp_path, db_name):
db_path = tmp_path / db_name
db.register(db_path, "https://alink", 1)
with pytest.raises(AssertionError):
db.write(db_path, range(7))
with pytest.raises(AssertionError):
db.append(db_path, range(7))
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
first_rev = get_rev(hg, date_from)
commits = hg_log(hg, first_rev)
commits_num = len(commits)
hg.close()
# Total previous number of commits by the author.
total_commits_by_author = defaultdict(int)
# Previous commits by the author, in a 90 days window.
commits_by_author = defaultdict(list)
global author_experience
global author_experience_90_days
for commit in commits:
author_experience[commit] = total_commits_by_author[commit.author]
# We don't want to consider backed out commits when calculating author/reviewer experience.
if not commit.ever_backedout:
total_commits_by_author[commit.author] += 1
# Keep only the previous commits from a window of 90 days in the commits_by_author map.
cut = None
for i, prev_commit in enumerate(commits_by_author[commit.author]):
if (commit.date - prev_commit.date).days <= 90:
break
cut = i
if cut is not None:
commits_by_author[commit.author] = commits_by_author[
commit.author][cut + 1:]
author_experience_90_days[commit] = len(
commits_by_author[commit.author])
if not commit.ever_backedout:
commits_by_author[commit.author].append(commit)
global COMPONENTS
r = requests.get(
'https://index.taskcluster.net/v1/task/gecko.v2.mozilla-central.latest.source.source-bugzilla-info/artifacts/public/components.json'
)
r.raise_for_status()
COMPONENTS = r.json()
print(f'Mining commits using {multiprocessing.cpu_count()} processes...')
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def test_append(mock_db, db_format, db_compression):
db_path = mock_db(db_format, db_compression)
db.write(db_path, range(1, 4))
assert list(db.read(db_path)) == [1, 2, 3]
db.append(db_path, range(4, 8))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7]
def test_append(mock_db, db_format, db_compression):
db_path = mock_db(db_format, db_compression)
db.write(db_path, range(1, 4))
assert list(db.read(db_path)) == [1, 2, 3]
db.append(db_path, range(4, 8))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7]
def test_exists_db(tmp_path):
db_path = tmp_path / "prova.json"
db.register(db_path, "https://alink", 1)
assert not db.exists(db_path)
db.write(db_path, range(7))
assert db.exists(db_path)
def test_delete(mock_db, db_format, db_compression):
db_path = mock_db(db_format, db_compression)
db.write(db_path, range(1, 9))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7, 8]
db.delete(db_path, lambda x: x == 4)
assert list(db.read(db_path)) == [1, 2, 3, 5, 6, 7, 8]
def test_delete(mock_db, db_format, db_compression):
db_path = mock_db(db_format, db_compression)
db.write(db_path, range(1, 9))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7, 8]
db.delete(db_path, lambda x: x == 4)
assert list(db.read(db_path)) == [1, 2, 3, 5, 6, 7, 8]
def update_commits() -> None:
commits = list(
get_commits(include_no_bug=True,
include_backouts=True,
include_ignored=True))
# Add coverage information for previous commits too.
# We need to do this because coverage information is sometimes slow to come in.
set_commit_coverage(commits)
db.write(COMMITS_DB, commits)
def test_unregistered_db(tmp_path):
db_path = tmp_path / "prova.json"
with pytest.raises(AssertionError):
list(db.read(db_path))
with pytest.raises(AssertionError):
db.write(db_path, range(7))
with pytest.raises(AssertionError):
db.append(db_path, range(7))
def download_commits(repo_dir):
commits = hg_log(repo_dir)
commits_num = len(commits)
print(f'Mining commits using {multiprocessing.cpu_count()} processes...')
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def test_unregistered_db(tmp_path):
db_path = tmp_path / "prova.json"
with pytest.raises(AssertionError):
list(db.read(db_path))
with pytest.raises(AssertionError):
db.write(db_path, range(7))
with pytest.raises(AssertionError):
db.append(db_path, range(7))
def test_append_compressed(tmp_path):
db_path = tmp_path / 'prova.json.gz'
db.register(db_path, 'https://alink', 1)
db.write(db_path, range(1, 4))
assert list(db.read(db_path)) == [1, 2, 3]
db.append(db_path, range(4, 8))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7]
def download_commits(repo_dir):
hg = hglib.open(repo_dir)
commits = hg.log()
hg.close()
commits = (tuple(commit) for commit in commits)
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=256)
db.write(COMMITS_DB, commits)
def test_delete_compressed(tmp_path):
db_path = tmp_path / 'prova.json.gz'
print(db_path)
db.register(db_path, 'https://alink', 1)
db.write(db_path, range(1, 9))
assert list(db.read(db_path)) == [1, 2, 3, 4, 5, 6, 7, 8]
db.delete(db_path, lambda x: x == 4)
assert list(db.read(db_path)) == [1, 2, 3, 5, 6, 7, 8]
def update_commits() -> None:
if not os.path.exists("data/coverage_mapping.lmdb"):
logger.info("Downloading commit->coverage mapping...")
download_coverage_mapping()
commits = list(
get_commits(include_no_bug=True,
include_backouts=True,
include_ignored=True))
# Add coverage information for previous commits too.
# We need to do this because coverage information is sometimes slow to come in.
set_commit_coverage(commits)
db.write(COMMITS_DB, commits)
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
revs = get_revs(hg, date_from)
commits_num = len(revs)
assert commits_num > 0, 'There should definitely be more than 0 commits, something is wrong'
hg.close()
processes = multiprocessing.cpu_count()
print(f'Mining {commits_num} commits using {processes} processes...')
CHUNK_SIZE = 256
revs_groups = [
revs[i:(i + CHUNK_SIZE)] for i in range(0, len(revs), CHUNK_SIZE)
]
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_hg_log, revs_groups, chunksize=20)
commits = tqdm(commits, total=len(revs_groups))
commits = list(itertools.chain.from_iterable(commits))
# Don't analyze backouts.
backouts = set(commit.backedoutby for commit in commits
if commit.backedoutby != b'')
commits = [commit for commit in commits if commit.node not in backouts]
# Don't analyze commits that are not linked to a bug.
commits = [commit for commit in commits if commit.bug != b'']
# Skip commits which are in .hg-annotate-ignore-revs (mostly consisting of very
# large and not meaningful formatting changes).
with open(os.path.join(repo_dir, '.hg-annotate-ignore-revs'), 'r') as f:
ignore_revs = set(l[:40].encode('utf-8') for l in f)
commits = [commit for commit in commits if commit.node not in ignore_revs]
commits_num = len(commits)
print(f'Analyzing {commits_num} patches...')
# Total previous number of commits by the author.
total_commits_by_author = defaultdict(int)
# Previous commits by the author, in a 90 days window.
commits_by_author = defaultdict(list)
global author_experience
global author_experience_90_days
for commit in commits:
author_experience[commit] = total_commits_by_author[commit.author]
# We don't want to consider backed out commits when calculating author/reviewer experience.
if not commit.backedoutby:
total_commits_by_author[commit.author] += 1
# Keep only the previous commits from a window of 90 days in the commits_by_author map.
cut = None
for i, prev_commit in enumerate(commits_by_author[commit.author]):
if (commit.date - prev_commit.date).days <= 90:
break
cut = i
if cut is not None:
commits_by_author[commit.author] = commits_by_author[
commit.author][cut + 1:]
author_experience_90_days[commit] = len(
commits_by_author[commit.author])
if not commit.backedoutby:
commits_by_author[commit.author].append(commit)
global COMPONENTS
r = requests.get(
'https://index.taskcluster.net/v1/task/gecko.v2.mozilla-central.latest.source.source-bugzilla-info/artifacts/public/components.json'
)
r.raise_for_status()
COMPONENTS = r.json()
print(f'Mining commits using {multiprocessing.cpu_count()} processes...')
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
revs = get_revs(hg)
assert (
len(revs) > 0
), "There should definitely be more than 0 commits, something is wrong"
hg.close()
# Skip commits which are in .hg-annotate-ignore-revs (mostly consisting of very
# large and not meaningful formatting changes).
with open(os.path.join(repo_dir, ".hg-annotate-ignore-revs"), "rb") as f:
ignore_revs = set(l[:40] for l in f)
revs = [rev for rev in revs if rev not in ignore_revs]
processes = multiprocessing.cpu_count()
print(f"Mining {len(revs)} commits using {processes} processes...")
CHUNK_SIZE = 256
revs_groups = [revs[i : (i + CHUNK_SIZE)] for i in range(0, len(revs), CHUNK_SIZE)]
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir,)
) as executor:
commits = executor.map(_hg_log, revs_groups, chunksize=20)
commits = tqdm(commits, total=len(revs_groups))
commits = list(itertools.chain.from_iterable(commits))
# Don't analyze backouts.
backouts = set(commit.backedoutby for commit in commits if commit.backedoutby != "")
commits = [commit for commit in commits if commit.node not in backouts]
# Don't analyze commits that are not linked to a bug.
commits = [commit for commit in commits if commit.bug != b""]
print("Downloading file->component mapping...")
global path_to_component
r = requests.get(
"https://index.taskcluster.net/v1/task/gecko.v2.mozilla-central.latest.source.source-bugzilla-info/artifacts/public/components.json"
)
r.raise_for_status()
path_to_component = r.json()
path_to_component = {
path: "::".join(component) for path, component in path_to_component.items()
}
calculate_experiences(commits)
# Exclude commits outside the range we care about.
commits = [commit for commit in commits if commit.pushdate > date_from]
commits_num = len(commits)
print(f"Mining {commits_num} commits using {processes} processes...")
global rs_parsepatch
import rs_parsepatch
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir,)
) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
revs = get_revs(hg, date_from)
commits_num = len(revs)
assert (
commits_num > 0
), "There should definitely be more than 0 commits, something is wrong"
hg.close()
processes = multiprocessing.cpu_count()
print(f"Mining {commits_num} commits using {processes} processes...")
CHUNK_SIZE = 256
revs_groups = [
revs[i:(i + CHUNK_SIZE)] for i in range(0, len(revs), CHUNK_SIZE)
]
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_hg_log, revs_groups, chunksize=20)
commits = tqdm(commits, total=len(revs_groups))
commits = list(itertools.chain.from_iterable(commits))
# Don't analyze backouts.
backouts = set(commit.backedoutby for commit in commits
if commit.backedoutby != b"")
commits = [commit for commit in commits if commit.node not in backouts]
# Don't analyze commits that are not linked to a bug.
commits = [commit for commit in commits if commit.bug != b""]
# Skip commits which are in .hg-annotate-ignore-revs (mostly consisting of very
# large and not meaningful formatting changes).
with open(os.path.join(repo_dir, ".hg-annotate-ignore-revs"), "r") as f:
ignore_revs = set(l[:40].encode("utf-8") for l in f)
commits = [commit for commit in commits if commit.node not in ignore_revs]
commits_num = len(commits)
print(f"Analyzing {commits_num} patches...")
# Total previous number of commits by the author.
total_commits_by_author = defaultdict(int)
# Previous commits by the author, in a 90 days window.
commits_by_author = defaultdict(list)
global author_experience
global author_experience_90_days
for commit in commits:
author_experience[commit.node] = total_commits_by_author[commit.author]
# We don't want to consider backed out commits when calculating author/reviewer experience.
if not commit.backedoutby:
total_commits_by_author[commit.author] += 1
# Keep only the previous commits from a window of 90 days in the commits_by_author map.
cut = None
for i, prev_commit in enumerate(commits_by_author[commit.author]):
if (commit.date - prev_commit.date).days <= 90:
break
cut = i
if cut is not None:
commits_by_author[commit.author] = commits_by_author[
commit.author][cut + 1:]
author_experience_90_days[commit.node] = len(
commits_by_author[commit.author])
if not commit.backedoutby:
commits_by_author[commit.author].append(commit)
global path_to_component
r = requests.get(
"https://index.taskcluster.net/v1/task/gecko.v2.mozilla-central.latest.source.source-bugzilla-info/artifacts/public/components.json"
)
r.raise_for_status()
path_to_component = r.json()
path_to_component = {
path: "::".join(component)
for path, component in path_to_component.items()
}
global components_touched_prev
global components_touched_prev_90_days
global files_touched_prev
global files_touched_prev_90_days
components_touched = defaultdict(int)
files_touched = defaultdict(int)
prev_commits_90_days = []
for commit in commits:
components = set(path_to_component[path] for path in commit.files
if path in path_to_component)
for component in components:
components_touched_prev[
commit.node] += components_touched[component]
components_touched[component] += 1
for path in commit.files:
files_touched_prev[commit.node] += files_touched[path]
files_touched[path] += 1
if len(commit.file_copies) > 0:
for orig, copied in commit.file_copies.items():
if orig in path_to_component and copied in path_to_component:
components_touched[path_to_component[
copied]] = components_touched[path_to_component[orig]]
files_touched[copied] = files_touched[orig]
for i, prev_commit in enumerate(prev_commits_90_days):
if (commit.date - prev_commit.date).days <= 90:
break
cut = i
if cut is not None:
prev_commits_90_days = prev_commits_90_days[cut + 1:]
components_touched_90_days = defaultdict(int)
files_touched_90_days = defaultdict(int)
for prev_commit in prev_commits_90_days:
components_prev = set(path_to_component[path]
for path in prev_commit.files
if path in path_to_component)
for component_prev in components_prev:
components_touched_90_days[component_prev] += 1
for path_prev in prev_commit.files:
files_touched_90_days[path_prev] += 1
if len(prev_commit.file_copies) > 0:
for orig, copied in prev_commit.file_copies.items():
if orig in path_to_component and copied in path_to_component:
components_touched_90_days[path_to_component[
copied]] = components_touched_90_days[
path_to_component[orig]]
files_touched_90_days[copied] = files_touched_90_days[orig]
components_touched_prev_90_days[commit.node] = sum(
components_touched_90_days[component] for component in components)
files_touched_prev_90_days[commit.node] = sum(
files_touched_90_days[path] for path in commit.files)
prev_commits_90_days.append(commit)
print(f"Mining commits using {multiprocessing.cpu_count()} processes...")
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir, )) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def generate_push_data(self, granularity: str, training_months: int,
reretrieve: int) -> None:
# We'll use the past training_months months only for training the model,
# but we use half training_months months more than that to calculate the
# failure statistics.
from_months = training_months + math.floor(training_months / 2)
# We use the actual date instead of 'today-X' aliases to avoid mozci caching
# this query.
from_date = datetime.utcnow() - relativedelta(months=from_months)
to_date = datetime.utcnow() - relativedelta(days=3)
pushes = mozci.push.make_push_objects(
from_date=from_date.strftime("%Y-%m-%d"),
to_date=to_date.strftime("%Y-%m-%d"),
branch="autoland",
)
if granularity == "label":
push_data_db = test_scheduling.PUSH_DATA_LABEL_DB
elif granularity == "group":
push_data_db = test_scheduling.PUSH_DATA_GROUP_DB
elif granularity == "config_group":
push_data_db = test_scheduling.PUSH_DATA_CONFIG_GROUP_DB
def cache_key(push: mozci.push.Push) -> str:
return f"push_data.{granularity}.{push.rev}"
def generate(
futures: List[concurrent.futures.Future],
) -> Generator[PushResult, None, None]:
nonlocal reretrieve
num_cached = 0
num_pushes = len(pushes)
for _ in tqdm(range(num_pushes)):
push = pushes.pop(0)
cached = futures.pop(0).result()
semaphore.release()
# Regenerating a large amount of data when we update the mozci regression detection
# algorithm is currently pretty slow, so we only regenerate a subset of pushes whenever we
# run.
if cached:
value, mozci_version = cached
# Regenerate results which were generated with an older version of mozci.
if reretrieve > 0 and mozci_version != MOZCI_VERSION:
cached = None
reretrieve -= 1
# Regenerate results which don't contain the fix revision.
elif len(value) != 5:
cached = None
if cached:
num_cached += 1
value, mozci_version = cached
assert len(value) == 5
yield value
else:
logger.info(
f"Analyzing {push.rev} at the {granularity} level...")
key = cache_key(push)
try:
if granularity == "label":
runnables = push.task_labels
elif granularity == "group":
runnables = push.group_summaries.keys()
elif granularity == "config_group":
runnables = push.config_group_summaries.keys()
value = (
tuple(push.revs),
push.backedoutby or push.bustage_fixed_by,
tuple(runnables),
tuple(push.get_possible_regressions(granularity)),
tuple(push.get_likely_regressions(granularity)),
)
mozci.config.cache.put(
key,
(value, MOZCI_VERSION),
mozci.config["cache"]["retention"],
)
assert len(value) == 5
yield value
except mozci.errors.MissingDataError:
logger.warning(
f"Tasks for push {push.rev} can't be found on ActiveData"
)
except Exception:
traceback.print_exc()
logger.info(
f"{num_cached} pushes were already cached out of {num_pushes}")
semaphore = threading.BoundedSemaphore(256)
def retrieve_from_cache(push):
semaphore.acquire()
return mozci.config.cache.get(cache_key(push))
with concurrent.futures.ThreadPoolExecutor() as executor:
futures = [
executor.submit(retrieve_from_cache, push) for push in pushes
]
try:
db.write(push_data_db, generate(futures))
except Exception:
for f in futures:
f.cancel()
try:
semaphore.release()
except ValueError:
continue
raise
zstd_compress(push_data_db)
def generate_push_data(self, pushes: List[mozci.push.Push],
granularity: str) -> None:
# We keep in the cache the fact that we failed to analyze a push for 10
# days, so if we re-run often we don't retry the same pushes many times.
MISSING_CACHE_RETENTION = 10 * 24 * 60
from_date = get_from_date(granularity)
pushes = [
push for push in pushes
if datetime.utcfromtimestamp(push.date) >= from_date
]
if granularity == "label":
push_data_db = test_scheduling.PUSH_DATA_LABEL_DB
elif granularity == "group":
push_data_db = test_scheduling.PUSH_DATA_GROUP_DB
elif granularity == "config_group":
push_data_db = test_scheduling.PUSH_DATA_CONFIG_GROUP_DB
def cache_key(push: mozci.push.Push) -> str:
return f"push_data.{granularity}.{push.rev}"
def generate(executor) -> Generator[PushResult, None, None]:
num_cached = 0
num_pushes = len(pushes)
# Regenerating a large amount of data when we update the mozci regression detection
# algorithm is currently pretty slow, so we only regenerate 1000 pushes whenever we
# run.
to_regenerate = 1000
semaphore = threading.BoundedSemaphore(256)
def retrieve_from_cache(push):
semaphore.acquire()
return adr.config.cache.get(cache_key(push))
futures = tuple(
executor.submit(retrieve_from_cache, push) for push in pushes)
for push, future in zip(tqdm(pushes), futures):
exc = future.exception()
if exc is not None:
logger.info(f"Exception {exc} while getting {push.rev}")
for f in futures:
f.cancel()
cached = future.result()
semaphore.release()
if cached and to_regenerate > 0:
value, mozci_version = cached
# Regenerate results which were generated when we were not cleaning
# up WPT groups.
if any(runnable.startswith("/") for runnable in value[1]):
cached = None
to_regenerate -= 1
"""# Regenerate results which were generated with an older version of mozci.
elif mozci_version != MOZCI_VERSION and to_regenerate > 0:
cached = None
to_regenerate -= 1"""
if cached is not None:
num_cached += 1
if cached:
value, mozci_version = cached
yield value
else:
logger.info(
f"Analyzing {push.rev} at the {granularity} level...")
key = cache_key(push)
try:
if granularity == "label":
runnables = push.task_labels
elif granularity == "group":
runnables = push.group_summaries.keys()
elif granularity == "config_group":
runnables = push.config_group_summaries.keys()
value = (
push.revs,
tuple(runnables),
tuple(push.get_possible_regressions(granularity)),
tuple(push.get_likely_regressions(granularity)),
)
adr.config.cache.put(
key,
(value, MOZCI_VERSION),
adr.config["cache"]["retention"],
)
yield value
except adr.errors.MissingDataError:
logger.warning(
f"Tasks for push {push.rev} can't be found on ActiveData"
)
adr.config.cache.put(key, (), MISSING_CACHE_RETENTION)
except Exception:
traceback.print_exc()
adr.config.cache.put(key, (), MISSING_CACHE_RETENTION)
logger.info(
f"{num_cached} pushes were already cached out of {num_pushes}")
with concurrent.futures.ThreadPoolExecutor() as executor:
db.write(push_data_db, generate(executor))
zstd_compress(push_data_db)
def generate_push_data(self, granularity: str) -> None:
# We'll use the past TRAINING_MONTHS months only for training the model,
# but we use half TRAINING_MONTHS months more than that to calculate the
# failure statistics.
from_months = TRAINING_MONTHS[granularity] + math.floor(
TRAINING_MONTHS[granularity] / 2
)
# We use the actual date instead of 'today-X' aliases to avoid adr caching
# this query.
from_date = datetime.utcnow() - relativedelta(months=from_months)
to_date = datetime.utcnow() - relativedelta(days=3)
pushes = mozci.push.make_push_objects(
from_date=from_date.strftime("%Y-%m-%d"),
to_date=to_date.strftime("%Y-%m-%d"),
branch="autoland",
)
if granularity == "label":
push_data_db = test_scheduling.PUSH_DATA_LABEL_DB
elif granularity == "group":
push_data_db = test_scheduling.PUSH_DATA_GROUP_DB
elif granularity == "config_group":
push_data_db = test_scheduling.PUSH_DATA_CONFIG_GROUP_DB
def cache_key(push: mozci.push.Push) -> str:
return f"push_data.{granularity}.{push.rev}"
def generate(
futures: List[concurrent.futures.Future],
) -> Generator[PushResult, None, None]:
num_cached = 0
num_pushes = len(pushes)
# Regenerating a large amount of data when we update the mozci regression detection
# algorithm is currently pretty slow, so we only regenerate a subset of pushes whenever we
# run.
to_regenerate = int(os.environ.get("OLD_RESULTS_TO_REGENERATE", 0))
for _ in tqdm(range(num_pushes)):
push = pushes.pop(0)
cached = futures.pop(0).result()
semaphore.release()
if cached and to_regenerate > 0:
value, mozci_version = cached
# Regenerate results which were generated when we were not cleaning
# up WPT groups.
if granularity == "group" and any(
runnable.startswith("/") for runnable in value[1]
):
cached = None
to_regenerate -= 1
# Regenerate results which were generated when we didn't get a correct
# configuration for test-verify tasks.
elif granularity == "config_group" and any(
"test-verify" in runnable[0] for runnable in value[1]
):
cached = None
to_regenerate -= 1
# Regenerate results which were generated with an older version of mozci.
elif mozci_version != MOZCI_VERSION:
cached = None
to_regenerate -= 1
if cached:
num_cached += 1
value, mozci_version = cached
yield value
else:
logger.info(f"Analyzing {push.rev} at the {granularity} level...")
key = cache_key(push)
try:
if granularity == "label":
runnables = push.task_labels
elif granularity == "group":
runnables = push.group_summaries.keys()
elif granularity == "config_group":
runnables = push.config_group_summaries.keys()
value = (
push.revs,
tuple(runnables),
tuple(push.get_possible_regressions(granularity)),
tuple(push.get_likely_regressions(granularity)),
)
adr.config.cache.put(
key,
(value, MOZCI_VERSION),
adr.config["cache"]["retention"],
)
yield value
except adr.errors.MissingDataError:
logger.warning(
f"Tasks for push {push.rev} can't be found on ActiveData"
)
except Exception:
traceback.print_exc()
logger.info(f"{num_cached} pushes were already cached out of {num_pushes}")
semaphore = threading.BoundedSemaphore(256)
def retrieve_from_cache(push):
semaphore.acquire()
return adr.config.cache.get(cache_key(push))
with concurrent.futures.ThreadPoolExecutor() as executor:
futures = [executor.submit(retrieve_from_cache, push) for push in pushes]
try:
db.write(push_data_db, generate(futures))
except Exception:
for f in futures:
f.cancel()
try:
semaphore.release()
except ValueError:
continue
raise
zstd_compress(push_data_db)
def retrieve_test_scheduling_history(self):
os.makedirs("data", exist_ok=True)
# Download previous cache.
cache_path = os.path.abspath("data/adr_cache")
if not os.path.exists(cache_path):
try:
download_check_etag(URL, "adr_cache.tar.xz")
with tarfile.open("adr_cache.tar.xz", "r:xz") as tar:
tar.extractall()
assert os.path.exists(
"data/adr_cache"), "Decompressed adr cache exists"
except requests.exceptions.HTTPError:
logger.info("The adr cache is not available yet")
# Setup adr cache configuration.
os.makedirs(os.path.expanduser("~/.config/adr"), exist_ok=True)
with open(os.path.expanduser("~/.config/adr/config.toml"), "w") as f:
f.write(f"""[adr.cache.stores]
file = {{ driver = "file", path = "{cache_path}" }}
""")
# Get the commits DB.
if db.is_old_version(
repository.COMMITS_DB) or not db.exists(repository.COMMITS_DB):
db.download(repository.COMMITS_DB, force=True)
# We'll use the past TRAINING_MONTHS months only for training the model,
# but we use 3 months more than that to calculate the failure statistics.
subprocess.run(
[
"run-adr",
"ahal/ci-recipes",
"recipe",
"-o",
os.path.abspath("push_data.json"),
"-f",
"json",
"push_data",
"--",
"--from",
f"today-{TRAINING_MONTHS + 3}month",
"--to",
"today-2day",
"--branch",
"autoland",
],
check=True,
stdout=subprocess.
DEVNULL, # Redirect to /dev/null, as the logs are too big otherwise.
)
HISTORY_DATE_START = datetime.now() - relativedelta(
months=TRAINING_MONTHS)
with open("push_data.json", "r") as f:
data = json.load(f)
push_data = {}
for row in data[1:]:
# Revision -> (all tasks, possible regressions, likely regressions)
push_data[row[0]] = (row[1], row[2], row[3])
HISTORICAL_TIMESPAN = 56
past_failures = {}
def get_past_failures(task, push_num):
if task not in past_failures:
past_failures[task] = repository.exp_queue(
push_num, HISTORICAL_TIMESPAN + 1, 0)
return past_failures[task][push_num]
def generate_data():
commits_with_data = set()
saved_nodes = set()
push_num = 0
for commit_data in tqdm(repository.get_commits()):
node = commit_data["node"]
if node not in push_data:
continue
commits_with_data.add(node)
commit_push_data = push_data[node]
for task in commit_push_data[0]:
if not any(task.startswith(j) for j in JOBS_TO_CONSIDER):
continue
total_failures = get_past_failures(task, push_num)
past_7_pushes_failures = total_failures - get_past_failures(
task, push_num - 7)
past_14_pushes_failures = total_failures - get_past_failures(
task, push_num - 14)
past_28_pushes_failures = total_failures - get_past_failures(
task, push_num - 28)
past_56_pushes_failures = total_failures - get_past_failures(
task, push_num - 56)
pushdate = dateutil.parser.parse(commit_data["pushdate"])
if pushdate > HISTORY_DATE_START:
saved_nodes.add(node)
yield {
"rev": node,
"name": task,
"failures": total_failures,
"failures_past_7_pushes": past_7_pushes_failures,
"failures_past_14_pushes": past_14_pushes_failures,
"failures_past_28_pushes": past_28_pushes_failures,
"failures_past_56_pushes": past_56_pushes_failures,
"is_possible_regression": task
in commit_push_data[1],
"is_likely_regression": task
in commit_push_data[2],
}
if task in commit_push_data[1] or task in commit_push_data[
2]:
past_failures[task][push_num] = total_failures + 1
push_num += 1
logger.info(f"push data nodes: {len(push_data)}")
logger.info(
f"commits linked to push data: {len(commits_with_data)}")
logger.info(f"saved push data nodes: {len(saved_nodes)}")
db.write(test_scheduling.TEST_SCHEDULING_DB, generate_data())
zstd_compress(test_scheduling.TEST_SCHEDULING_DB)
with tarfile.open("data/adr_cache.tar.xz", "w:xz") as tar:
tar.add("data/adr_cache")
def download_commits(repo_dir, date_from):
hg = hglib.open(repo_dir)
revs = get_revs(hg)
assert (
len(revs) > 0
), "There should definitely be more than 0 commits, something is wrong"
hg.close()
processes = multiprocessing.cpu_count()
print(f"Mining {len(revs)} commits using {processes} processes...")
CHUNK_SIZE = 256
revs_groups = [revs[i : (i + CHUNK_SIZE)] for i in range(0, len(revs), CHUNK_SIZE)]
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir,)
) as executor:
commits = executor.map(_hg_log, revs_groups, chunksize=20)
commits = tqdm(commits, total=len(revs_groups))
commits = list(itertools.chain.from_iterable(commits))
print("Downloading file->component mapping...")
global path_to_component
r = requests.get(
"https://index.taskcluster.net/v1/task/gecko.v2.mozilla-central.latest.source.source-bugzilla-info/artifacts/public/components.json"
)
r.raise_for_status()
path_to_component = r.json()
path_to_component = {
path: "::".join(component) for path, component in path_to_component.items()
}
commits_to_ignore = get_commits_to_ignore(repo_dir, commits)
print(f"{len(commits_to_ignore)} commits to ignore")
calculate_experiences(commits, commits_to_ignore)
# Exclude commits to ignore.
commits = [commit for commit in commits if commit not in commits_to_ignore]
# Exclude commits outside the range we care about.
commits = [commit for commit in commits if commit.pushdate > date_from]
commits_num = len(commits)
print(f"Mining {commits_num} commits using {processes} processes...")
global rs_parsepatch
import rs_parsepatch
with concurrent.futures.ProcessPoolExecutor(
initializer=_init, initargs=(repo_dir,)
) as executor:
commits = executor.map(_transform, commits, chunksize=64)
commits = tqdm(commits, total=commits_num)
db.write(COMMITS_DB, commits)
def generate_push_data(self, granularity: str) -> None:
# We keep in the cache the fact that we failed to analyze a push for 10
# days, so if we re-run often we don't retry the same pushes many times.
MISSING_CACHE_RETENTION = 10 * 24 * 60
# We'll use the past TRAINING_MONTHS months only for training the model,
# but we use half TRAINING_MONTHS months more than that to calculate the
# failure statistics.
from_months = TRAINING_MONTHS[granularity] + math.floor(
TRAINING_MONTHS[granularity] / 2
)
# We use the actual date instead of 'today-X' aliases to avoid adr caching
# this query.
from_date = datetime.utcnow() - relativedelta(months=from_months)
to_date = datetime.utcnow() - relativedelta(days=3)
pushes = mozci.push.make_push_objects(
from_date=from_date.strftime("%Y-%m-%d"),
to_date=to_date.strftime("%Y-%m-%d"),
branch="autoland",
)
if granularity == "label":
push_data_db = test_scheduling.PUSH_DATA_LABEL_DB
elif granularity == "group":
push_data_db = test_scheduling.PUSH_DATA_GROUP_DB
cache: Dict[mozci.push.Push, Tuple[PushResult, int]] = {}
def cache_key(push: mozci.push.Push) -> str:
return f"push_data.{granularity}.{push.rev}"
with concurrent.futures.ThreadPoolExecutor() as executor:
future_to_push = {
executor.submit(
lambda push: adr.config.cache.get(cache_key(push)), push
): push
for push in pushes
}
for future in tqdm(
concurrent.futures.as_completed(future_to_push),
total=len(future_to_push),
):
push = future_to_push[future]
exc = future.exception()
if exc is not None:
logger.info(f"Exception {exc} while getting {push.rev}")
for f in future_to_push.keys():
f.cancel()
cache[push] = future.result()
# Regenerating a large amount of data when we update the mozci regression detection
# algorithm is currently pretty slow, so we only regenerate 1000 pushes whenever we
# run.
"""to_regenerate = 0
for push in pushes[::-1]:
cached = cache[push]
if not cached:
continue
value, mozci_version = cached
if mozci_version != MOZCI_VERSION and to_regenerate < 1000:
cache[push] = None
to_regenerate += 1"""
to_regenerate = 0
for push in pushes[::-1]:
cached = cache[push]
if not cached:
continue
if to_regenerate < 1000:
del cache[push]
adr.config.cache.put(push.push_uuid, {}, 0)
to_regenerate += 1
def generate() -> Generator[PushResult, None, None]:
num_cached = 0
for push in tqdm(pushes):
key = cache_key(push)
if push in cache and cache[push] is not None:
num_cached += 1
cached = cache[push]
if cached:
value, mozci_version = cached
yield value
else:
logger.info(f"Analyzing {push.rev} at the {granularity} level...")
try:
if granularity == "label":
runnables = push.task_labels
elif granularity == "group":
runnables = push.group_summaries.keys()
value = (
push.revs,
list(runnables),
list(push.get_possible_regressions(granularity)),
list(push.get_likely_regressions(granularity)),
)
adr.config.cache.put(
key,
(value, MOZCI_VERSION),
adr.config["cache"]["retention"],
)
yield value
except adr.errors.MissingDataError:
logger.warning(
f"Tasks for push {push.rev} can't be found on ActiveData"
)
adr.config.cache.put(key, (), MISSING_CACHE_RETENTION)
except Exception:
traceback.print_exc()
adr.config.cache.put(key, (), MISSING_CACHE_RETENTION)
logger.info(f"{num_cached} pushes were already cached out of {len(pushes)}")
db.write(push_data_db, generate())
zstd_compress(push_data_db)
def generate_push_data(
self, pushes: Tuple[mozci.push.Push, ...], granularity: str
) -> None:
from_date = get_from_date(granularity)
pushes = tuple(
push for push in pushes if datetime.utcfromtimestamp(push.date) >= from_date
)
if granularity == "label":
push_data_db = test_scheduling.PUSH_DATA_LABEL_DB
elif granularity == "group":
push_data_db = test_scheduling.PUSH_DATA_GROUP_DB
elif granularity == "config_group":
push_data_db = test_scheduling.PUSH_DATA_CONFIG_GROUP_DB
def cache_key(push: mozci.push.Push) -> str:
return f"push_data.{granularity}.{push.rev}"
def generate(
futures: List[concurrent.futures.Future],
) -> Generator[PushResult, None, None]:
num_cached = 0
num_pushes = len(pushes)
# Regenerating a large amount of data when we update the mozci regression detection
# algorithm is currently pretty slow, so we only regenerate 1000 pushes whenever we
# run.
to_regenerate = 1000
for push in tqdm(pushes):
cached = futures.pop(0).result()
semaphore.release()
if cached and to_regenerate > 0:
value, mozci_version = cached
# Regenerate results which were generated when we were not cleaning
# up WPT groups.
if granularity == "group" and any(
runnable.startswith("/") for runnable in value[1]
):
cached = None
to_regenerate -= 1
# Regenerate results which were generated when we didn't get a correct
# configuration for test-verify tasks.
elif granularity == "config_group" and any(
"test-verify" in runnable[0] for runnable in value[1]
):
cached = None
to_regenerate -= 1
"""# Regenerate results which were generated with an older version of mozci.
elif mozci_version != MOZCI_VERSION:
cached = None
to_regenerate -= 1"""
if cached:
num_cached += 1
value, mozci_version = cached
yield value
else:
logger.info(f"Analyzing {push.rev} at the {granularity} level...")
key = cache_key(push)
try:
if granularity == "label":
runnables = push.task_labels
elif granularity == "group":
runnables = push.group_summaries.keys()
elif granularity == "config_group":
runnables = push.config_group_summaries.keys()
value = (
push.revs,
tuple(runnables),
tuple(push.get_possible_regressions(granularity)),
tuple(push.get_likely_regressions(granularity)),
)
adr.config.cache.put(
key,
(value, MOZCI_VERSION),
adr.config["cache"]["retention"],
)
yield value
except adr.errors.MissingDataError:
logger.warning(
f"Tasks for push {push.rev} can't be found on ActiveData"
)
except Exception:
traceback.print_exc()
logger.info(f"{num_cached} pushes were already cached out of {num_pushes}")
semaphore = threading.BoundedSemaphore(256)
def retrieve_from_cache(push):
semaphore.acquire()
return adr.config.cache.get(cache_key(push))
with concurrent.futures.ThreadPoolExecutor() as executor:
futures = [executor.submit(retrieve_from_cache, push) for push in pushes]
try:
db.write(push_data_db, generate(futures))
except Exception:
for f in futures:
f.cancel()
try:
semaphore.release()
except ValueError:
continue
raise
zstd_compress(push_data_db)
